Predicting EPL results with Elo rating, league simulation and machine learning prediction model

• Main Author: Ching, Kai Teng
• Other Authors: Yong Ee Hou
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Science::Physics::Descriptive and experimental mechanics
• Online Access: https://hdl.handle.net/10356/156903
• Institution: Nanyang Technological University

There are many techniques to predict the outcome of professional football matches whether by goal score or team strength or even past results. However, there is a lot of random element involved in a game of football, goal scores may be the results of better luck or a keeper’s error. Team strength can be handicapped when the key player is injured or is forced to take an international break in favor of playing for his country. Even past results may be skewed as there are home teams and away teams in football matches and past results sometimes may be skewed towards having a weaker opponent. The main objective of this project is to explore different techniques that are logical to try and predict the outcome and scores of football matches that happen in the English Premier League, with Machine Learning , Elo rating and League simulation.The different techniques and hypotheses will be tested and the accuracy of the results will be tested for all different techniques to see which of the system works the best and in which types of conditions. In this thesis, for the League simulation a team overall rating from each player will be generated with a calculation of a team’s offensive and defensive ratings which will generate a set of results. For Elo rating, the system will be based on predicting the win and loss of the matches from the team’s standings. Lastly for machine learning, the SVM model will be based on goals which will generate the league table for win and loss while the logistic regression will be based on Elo to predict outcome, with the higher accuracy AI used in the analysis. The different prediction models will be compared against each other to see which is best.